Enabling LLM Knowledge Analysis via Extensive Materialization

TL;DR

This paper introduces a comprehensive methodology for analyzing LLMs' factual knowledge by extensive materialization through recursive querying, resulting in a large knowledge base that enables detailed insights into the scope, accuracy, and biases of LLMs.

Contribution

It presents a novel recursive querying approach to fully materialize LLM knowledge, overcoming previous limitations of small sample analysis, and provides the GPTKB knowledge base as a resource.

Findings

GPTKB contains 101 million triples for 2.9 million entities.

Analysis reveals insights into GPT-4o-mini's scale, accuracy, and biases.

The methodology enables comprehensive understanding of LLM knowledge structures.

Abstract

Large language models (LLMs) have majorly advanced NLP and AI, and next to their ability to perform a wide range of procedural tasks, a major success factor is their internalized factual knowledge. Since Petroni et al. (2019), analyzing this knowledge has gained attention. However, most approaches investigate one question at a time via modest-sized pre-defined samples, introducing an ``availability bias'' (Tversky&Kahnemann, 1973) that prevents the analysis of knowledge (or beliefs) of LLMs beyond the experimenter's predisposition. To address this challenge, we propose a novel methodology to comprehensively materialize an LLM's factual knowledge through recursive querying and result consolidation. Our approach is a milestone for LLM research, for the first time providing constructive insights into the scope and structure of LLM knowledge (or beliefs). As a prototype, we build GPTKB, a knowledge base (KB) comprising 101 million relational triples for over 2.9 million entities from GPT-4o-mini. We use GPTKB to exemplarily analyze GPT-4o-mini's factual knowledge in terms of scale, accuracy, bias, cutoff and consistency, at the same time. GPTKB is accessible at https://gptkb.org